Introduction

The purpose of the Behavioral Health Equity Index (BHEI) Technical Report is to summarize the methods and preliminary results for the BHEI. The BHEI is a descriptive, data-driven tool that allows users to explore differences in the underlying, or root causes, of behavioral health across neighborhoods in San Diego County. The index is constructed from over 30 indicators, organized into eight domains that map to five social determinants of behavioral health. Neighborhoods with higher BHEI scores are relatively less likely to have access to the resources, opportunities, and conditions that promote behavioral health than neighborhoods with lower BHEI scores. The BHEI may be used to inform culturally responsive, data-informed outreach efforts to previously underserved areas in San Diego County.

The BHEI is a product of the Community Experience Partnership (CEP), a joint initiative between County of San Diego Behavioral Health Services (BHS) and UC San Diego’s Child and Adolescent Services Research Centers (CASRC) and Health Services Research Center (HSRC). The goal of the CEP is to integrate data and community engagement to promote behavioral health equity in San Diego County.

Construction

The BHEI was developed using the R package COINr: Composite Indicator Construction and Analysis (v1.1.4; Becker; 2022). In line with standard practice (e.g., Becker, 2022, Nardo et al., 2008), construction included the following steps:

  1. Defining the theoretical framework
  2. Selecting the indicators
  3. Finalizing the conceptual model
  4. Defining the geographic units of analysis
  5. Creating the preliminary dataset
  6. Applying inclusion/exclusion criteria
  7. Imputing missing data
  8. Treating outliers
  9. Normalizing indicators
  10. Defining weights
  11. Aggregating and ranking indicators and domains.

Theoretical Framework

The BHEI is grounded in a social determinants of behavioral health framework. Specifically, the eight BHEI domains map to the Centers for Disease Control and Prevention’s Social Determinants of Health (SDOH) Framework (Healthy People 2030, 2022). This framework identifies five SDOH domains as:

  1. Economic stability
  2. Education quality and access
  3. Health care access and quality
  4. Neighborhood and built environment
  5. Social and community context.

In this framework, behavioral health is shaped by the environments in which people live, learn, work, and play. Inequities are caused by unequal access to resources and opportunities and are perpetuated through unfair systems, policies, and laws. These disparities disproportionately affect communities of color, sexual minorities, those living in poverty, and other historically underserved groups.

Understanding where, geographically, inequities exist in a community is a first step towards identifying and addressing the policies, laws, and services that create or exacerbate behavioral health inequities. In this way, the BHEI may help identify neighborhoods or regions that could benefit from behavioral health service enhancements or quality improvement efforts.

Indicator Selection

Candidate Indicators

Initially, the BHEI development team at UC San Diego conducted a literature review to identify a list of candidate indicators. Proposed indicators met the following criteria:

  1. Available at the census tract or zip code level
  2. Theoretically related to one of the social determinants of behavioral health, or a geographic predictor of behavioral health equity
  3. Taken from a reliable and valid data source

Community Workgroup

The Community Experience Committee Workgroup was formed in January 2022 to help review, edit, and finalize the BHEI indicators and to advise on the theoretical framework, conceptual model, and BHEI weighting methodology. The workgroup was composed of Subject Matter Experts (SMEs) including community experts, representatives from County of San Diego BHS, and UC San Diego researchers.

A web-based survey was developed and administered to gather member feedback about the proposed BHEI indicator list. Results were discussed during four working sessions conducted between January and April 2022. The following indicators were added to the candidate list as a result of these discussions:

  1. Preschool enrollment
  2. Routine dental visit (past 12 months)
  3. Routine checkup in (past 12 months)
  4. Voter participation
  5. Residential diversity
  6. Mean income at age 35
  7. Incarceration at age 35

Data sources were not available to capture all indicators that members identified as being important causes of behavioral health equity in San Diego County. Most notably, members were concerned that systemic racism and food access were not well reflected in existing data sources and that groups who may benefit from additional supports or services (e.g., people identifying as LGBTQ+) could not be identified.

Additionally, although workgroup members supported a SDOH theoretical framework, community experts were critical of the initial framing of the BHEI, which focused on unmet needs and community deficits. Members were concerned that the index was not strength based and could be stigmatizing to neighborhoods. Based on this feedback, the index was reframed as a descriptive tool to help understand how the underlying, or root causes, of behavioral health equity vary across neighborhoods in San Diego County.

Conceptual Model

Once the indicator list was finalized, indicators were organized into eight domains that mapped to the Centers for Disease Control’s (CDC) five Social Determinants of Health (SDOH).

The final model for the BHEI is presented in Figure 1. Additional information on the indicators that make up the BHEI, including the data sources, most recent year(s) the data were produced, and technical definitions, are included in Appendix A.

Figure 1: BHEI Conceptual Model

Zoom Image

Geographies

The BHEI is calculated at four geographic units of analysis:

  1. 2020 Census Tracts: Census tracts are geographic boundaries defined by the U.S. Census. Tracts typically have populations between 1,200 and 8,000 people and are often described as representing “neighborhoods.” The BHEI uses geographic boundaries established by the 2020 U.S. Census Bureau. After excluding nine census tracts that failed to meet the geographic inclusion criteria, 727 census tracts were assigned a BHEI ranking.

  2. 2020 Zip Code Tabulation Areas (ZCTAs): Zip code data are based on Zip Code Tabulation Areas (ZCTAs) defined by the 2020 U.S. Census. Because zip codes are not a reliable geography, the Census Bureau assigns ZCTAs using the most frequently occurring zip code in a small census area. In most cases zip codes are the same as ZCTAs, however, some addresses are assigned to a ZCTA that differs from their zip code. Additionally, zip codes representing very few addresses may not be represented by a ZCTA. Additionally, ZCTAs may cross San Diego County boundaries and large bodies of water and unpopulated land areas do not have ZCTAs. After excluding 19 ZCTAs that failed to meet the geographic inclusion criteria, 97 zip codes were assigned a BHEI ranking.

  3. Subregional Areas (SRAs): SRAs are aggregated census tracts defined by the San Diego Association of Governments (SANDAG). SRAs were originally derived from 1970s vintage census tracts and assigned the name of the largest city or community in the area. SRA boundaries have not changed over time. All 41 SRAs met the geographic inclusion criteria and were assigned a BHEI rank.

  4. Health and Human Services Agency (HHSA) Regions: There are six County of San Diego Board of Supervisor approved HHSA regions in San Diego County. HHSA regions are designed to assist with the organization and administration of health service delivery and operations. The Service Planning Tool aggregates SRAs to form HHSA geographies. All six HHSAs met the geographic inclusion criteria and were assigned a BHEI rank.

The boundaries for each of these geographic areas can be viewed in the BHEI Maps section of this report.

Pre-Processing

Indicator data were downloaded from the sources listed in Appendix A. All analyses were conducted in R (v4.1.2; R Core Team, 2021). Indicators sourced from the U.S. Census Bureau’s American Community Survey (ACS) were downloaded at the census tract and ZCTA geographies using the R package Tidycensus (v1.3.3; Walker & Herman, 2023).

With few exceptions, raw indicator data included numerators and denominators for each geographic unit, used percent as the measurement unit (i.e., as opposed to rates, counts, etc.), and were available at the census tract geography. Some indicators did not meet these standards and required additional data processing as described below.

Denominators and Numerators

Data sourced from PLACES and the Opportunity Atlas (Appendix A) did not provide numerator and denominator estimates. After confirming procedures with the CDC, numerators were estimated by multiplying the tract proportion for each indicator by the appropriate population universe (e.g., the estimated number of adults in each geographic area was used for PLACES data). To align with data sourced from the ACS, the denominators were based on ACS 2015-2019 tract level estimates. Although 5-year ACS data are imperfect estimates of population counts, recent research has demonstrated these estimates work relatively well for area-aggregate models and introduce less bias than commonly used machine-learning models (Nethery et al., 2021).

Rates for Discharged Clients

Counts of discharges from emergency departments and inpatient facilities with any mention of Mental and Behavioral Disorders (MBD) were sourced from the California Department of Health Care Access and Information (HCAI) and prepared by County of San Diego’s Behavioral Health Services Population Health Unit (HCAI, v. 04/23). Data were provided as counts at the zip code level based on the discharging client’s home address. To improve reliability and exceed small count suppression thresholds (e.g., non-zero counts of less than 15 were suppressed), the Population Health Unit aggregated data across six years (2016-2021).

Initially suppressed counts were imputed with a value of 6 (the average of all suppressed values). SANDAG’s Population and Housing Estimate crosswalk (SANDAG, v2021) was then used to reallocate the zip code counts to 2010 census tract geographies (crosswalks based on the 2020 geographic boundaries were not available at the time of analysis). Zip codes 92259, 92536, and 92672 crossed into neighboring counties and did not appear in the HCAI data, although they were represented in the SANDAG crosswalks.

There were 103,425 eligible discharge events from emergency department facilities and 57,449 from inpatient facilities where the discharging individual was experiencing homelessness and did not have a recorded home zip code. Rather than exclude these clients or impute the address of their discharging facility, a proportional allocation method was used to assign data to 2010 census tracts using proportions based on Point-in-Time counts. Specifically, UC San Diego’s Homelessness Hub produces a public dataset that assigns the 2018 Point-in-Time counts collected by the Regional Task Force on Homelessness (RTFH) to 2010 census tracts (UC San Diego’s Homelessness Hub, v03/2022). This file was used to proportionally allocate discharge data for clients experiencing homelessness. At the time of analysis, SANDAG population estimates did not account for individuals experiencing homelessness, which may introduce some bias in rate estimates.

Once all data were allocated to the 2010 census tract geographies, allocation to 2020 census tracts, ZCTAs, SRAs and HHSAs were conducted using the procedures outlined in the Geographic Reallocation section of this report. Finally, rates were calculated as the estimated count of discharged clients in each geographic area divided by the corresponding estimated population universe over the six-year period multiplied by 100,000.

Geographic Reallocation

Between each decennial census, the estimated number of residents living in a census designated area may change. Census geographies may be merged or divided at each decennial census depending on whether the population increased or decreased since the last census. At the time the BHEI was generated, only indicators that were sourced from the ACS were available in the redefined 2020 census tract and ZCTA geographies.

Non-ACS data were reallocated to the 2020 geographic boundaries using geographic equivalency files and established processing algorithms. The basic process involves merging an equivalency file for the target area to the source file, multiplying the numerators and denominators for each indicator by the equivalency weight, aggregating the new estimates to the target geography, and generating a new proportion. The process for data allocation using geographic equivalency files is described in detail by the Missouri Census Data Center.

For the purposes of the BHEI, the following geographic allocation files and procedures were used:

  1. 2010 census tract data were reallocated to 2020 census tracts using equivalency files generated by iPUMS National Historical Geographic Information System (NHGIS; Manson et al.; v2021).
  2. Census tract 2020 data were assigned to ZCTA 2020 geographies using equivalency files produced by the Missouri Census Data Center’s GEOCORR 2022: Geographic Correspondence Engine (Missouri Census Data Center, v2022).
  3. Census tract 2020 data were assigned to SRAs using the San Diego Association of Government’s (SANDAG) allocation files (received via personal email communication; v09/09/22).
  4. SRAs were aggregated to form the six HHSA geographies using crosswalks based on publicly available data presentations from County of San Diego, Health and Human Services Agency, Community Health Statistics Unit (CHSU, v2023).

Inclusion Criteria

In general, units with a high proportion of missing data, few unique values (for indicators), a high proportion of residents in group housing (for geographies), or large coefficients of variation were assumed to have diminished discriminatory power and considered unreliable predictors likely to bias the BHEI.

Thresholds to define “large” coefficients of variation are somewhat arbitrary. Esri considers an ACS estimate unreliable if the coefficient of variation is over 40% (ESRI, 2022). The same 40% threshold was used to identify “unreliable estimates” for BHEI indicators sourced from the ACS. Because margins or error were not available for indicators reallocated to 2020 geographies, reliability was only assessed for variables sourced from the ACS.

Indicator Criteria

Candidate BHEI indicators were only included in the BHEI if they met the following criteria:

  1. More than 90% of census tract estimates were non-missing (i.e., less than 10% missing)
  2. At least 15% of census tract estimates were unique
  3. The majority of ACS-based estimates were “reliable” (i.e., the coefficient of variation of the denominator was <40% for a majority of estimates)

Based on these criteria, the following 5 candidate indicators were excluded: percent of households with incomplete kitchens, percent of household with incomplete plumbing, allegations of child maltreatment, adult incarceration at age 35, and the percent of 3- and 4-year-olds enrolled in preschool. Thirty-one indicators met the inclusion criteria and were retained in the index.

Table 1 shows the criteria for each of the excluded indicators along with flags when a specific criterion was not met. Although it does not appear in the table, “low-income food insecurity” was also excluded due to a high proportion of suppressed/missing data (196/627 = 31%). This indicator was not publicly available and does not appear in the table below.

Table 1: Excluded BHEI indicators

Geographic Criteria

Geographies were included in the BHEI if:

  1. More than 90% of estimates were non-missing (i.e., less than 10% missing)
  2. The majority of ACS-based estimates were “reliable” (i.e., the coefficient of variation of the denominator was <40% for a majority of estimates)
  3. At least 65% of the population were not living in group quarters

Group quarters include “college residence halls, residential treatment centers, skilled nursing facilities, group homes, military barracks, correctional facilities for adults, and workers group living quarters and job corps centers” (U.S. Census Bureau, 2021). Because the characteristics and needs of individuals living in group quarters differ from those living in households, behavioral health equity for the group quarters population cannot be accurately measured by the BHEI. In San Diego County, several excluded geographies have high concentrations of active military living in barracks. Users should be aware that behavioral health outreach to these communities cannot be informed by the BHEI. Table 2 shows the count and percent of excluded geographies by geographic unit. No SRAs or HHSAs were excluded.

Table 2: Excluded geographies by geographic unit

Total
Exclude Include
Type
    Census Tract 9 (1.2%) 727 (99%) 736 (100%)
    HHSA 0 (0%) 6 (100%) 6 (100%)
    SRA 0 (0%) 41 (100%) 41 (100%)
    ZCTA 19 (16%) 97 (84%) 116 (100%)
Total 28 (3.1%) 871 (97%) 899 (100%)

Tables 3 and 4 show the criteria for each of the 28 excluded geographies. Group quarters data appear as missing when the ACS estimated total population for the area is zero.

Census Tracts

Table 3: Excluded census tracts

ZCTAs

Table 4: Excluded ZCTAs

Missing Data Imputation

Missing data were imputed by assigning the value from the next largest, non-missing geographic unit in which the area nested. For instance, missing census tract estimates were assigned the non-missing value from the SRA in which they nested. Table 5 summarizes the proportion of missing data that were imputed for each geographic unit and indicator. No data were imputed for ZCTAs or HHSAs.

Table 5: Missing data by indicator and geographic unit

Geography Total
Census Tract SRA
Indicator
    Homeowner cost burden 4 (80%) 1 (20%) 5 (100%)
    Outcomes in adulthood (income) 1 (100%) 0 (0%) 1 (100%)
    Proximity to offsite alcohol outlets  19 (100%) 0 (0%) 19 (100%)
    Renter cost burden 1 (100%) 0 (0%) 1 (100%)
    Single-parent households 1 (100%) 0 (0%) 1 (100%)
Total 26 (96%) 1 (3.7%) 27 (100%)

Data Treatment

Outlier data or heavily skewed distributions can cause normalized indicators to have a large proportion of very high or low scores. This may diminish an indicator’s discriminatory power and result in poorer statistical properties for aggregation (Becker, 2022).

Skewness and kurtosis were calculated to assess the normality of the indicator distributions at each geographic unit of analysis. Skewness measures the asymmetry of a distribution where a value of 0 suggests symmetry, a positive value indicates a long right tail, and a negative value indicates a long left tail. Kurtosis is a measure of peakedness where a positive kurtosis value indicates high peak distributions (i.e., leptokurtic) and a negative value indicates flat-topped curve distributions (i.e., platykurtic). Indicators are flagged as non-normal if they have an absolute skewness > 2 AND an absolute kurtosis > 3.5 (Kim, 2013).

Using these criteria, a small proportion of BHEI indicators were flagged as “non-normal.” To reduce the influence of outliers on the BHEI, Winsorization was applied to all indicators that failed to meet normality thresholds. Winsorization involves replacing the most outlying observations with the next highest retained point.

Non-normal indicators were detected at the Census Tract, ZCTA, and SRA geographies. Tables 6-8 show the skewness and kurtosis of non-normal values at each geographic unit before and after Winsorization, as well as the number of Winsorization points that were required to meet the normality threshold (i.e., the number of geographies that were treated). No HHSAs were treated.

Census Tract

Table 6: Results of data treatment by census tract

ZCTA

Table 7: Results of data treatment by ZCTA

SRA

Table 8: Results of data treatment by SRA

Normalization

Because the BHEI indicators use different measurement units (e.g., percentages, scores, dollars, rates, etc.), it is necessary to normalize indicators, or bring them onto the same scale prior to aggregation. Normalization was conducted in two stages. First, indicators that were negatively correlated to the BHEI concept were rescaled to be in the same direction as other indicators (e.g., to be positively correlated with greater inequity). The following indicators were rescaled:

Table 9: Rescaled indicators

Next, for each geographic unit of analysis, indicators were scaled by calculating the z-scores. The formula for calculating the z-score is given by:

\[ z = \frac{{x - \mu}}{{\sigma}} \]

where:

  • \(z\) is the z-score,

  • \(x\) is the observed value,

  • \(\mu\) is the mean of the population or sample, and

  • \(\sigma\) is the standard deviation of the population or sample.

Z-scores have the advantage of being less sensitive to outliers than other aggregation techniques like the “Min-Max” method (Becker, 2022).

Weighting

Each of the eight BHEI Domains were assumed to differentially impact behavioral health equity in San Diego County. Domains should, therefore, be weighted according to their relative importance towards determining behavioral health equity. The BHEI development team elected to use a participatory weighting process to determine domain weights for the BHEI. Participatory approaches, although less frequently used than other weighting schemes (Greco et al., 2019; Benjamin et al., 2013; Decancq & Lugo, 2013; Nardo et al., 2009), align well with the values and goals of the CEP (e.g., transparency and stakeholder engagement). Initially, a stated preference weights approach was considered as a methodology to achieve community representation. In this approach weights are based on the opinions of a representative sample of individuals (Benjamin et al., 2013; Decancq & Lugo, 2013;).

The Community Experience Committee Workgroup was convened, in part, to advise the engagement process around the community sample for the purposes of weighting. The development team received a strong recommendation from the workgroup not to survey community members for this purpose. Workgroup members noted a high level of survey fatigue within the community, that any future community surveys should be strengths based, and that a weighting survey could compromise opportunities for more valuable engagement around the CEP.

Based on this feedback, UC San Diego researchers, in consultation with BHS representatives, opted to limit the weighting sample to community experts, stakeholders, and technical experts. To collect weighting data from the sample of experts, the BHEI development team elected to use the budget allocation process (BAP), a participatory weighting method. In the BAP a group of expert decision makers distributed 100 points to the eight BHEI domains based on their perceived value or importance for achieving behavioral health equity in San Diego County. The scores were then averaged to calculate an “expert weight” (Greco et al., 2019; Decancq & Lugo, 2013; Nardo et al., 2009).

Four expert groups were selected to receive the BAP weighting survey:

  1. A focus group consisting of diverse community stakeholders, including program managers and directors representing local behavioral health agencies and advocacy groups;
  2. The Community Experience Committee Workgroup which included community experts, UC San Diego researchers and BHS representatives,
  3. The Behavioral Health Advisory Board (BHAB)
  4. The Behavioral Health Services Leadership Team

The BAP survey was developed and programmed into Qualtrics for administration. Twenty-three decision makers provided complete weighting responses.

To understand if community experts would weight domains differently from technical experts, mean scores were calculated for both categories of respondents (Table 10). Community representatives included Community Experience Committee Workgroup members that self-identified as “Community Partners” and individuals who attended the community focus group. Technical experts included representatives from BHS, UC San Diego, and BHAB. Differences between the scores were nominal (Table 10). Similarly, sensitivity analyses revealed that the weighting scheme (i.e., community only weights, technical expert weights, all representative weights) had very little effect on the final BHEI score. The BHEI was therefore weighted using mean scores based on responses from all 23 representatives.

Table 10: Domain weights by respondent type

Aggregation

Aggregation is the process of combining normalized indicators into a single summary score. An important consideration when selecting the aggregation method for an index is determining if a high score in one indicator should compensate for a low score in another. For instance, as Becker (2023) describes the arithmetic mean as a perfectly compensatory method because “with two indicators scaled between 0 and 10 and equal weighting, a unit with scores (0, 10) would be given the same score as a unit with scores (5, 5) both have a score of 5” (Becker, 2023).

In the case of the BHEI, the development team did not feel the assumption of compensability held since most, if not all, the indicators and domains could be seen as “essentials” (e.g., a high score on the “Economic Domain” cannot entirely compensate for a low score on the “Education Domain”). The weighted geometric mean was selected as the aggregation method because it is a less compensatory method than other techniques (i.e., the arithmetic mean).

The weighted geometric mean is calculated as:

\[\left( \prod_{{i=1}}^{{n}} x_i^{w_i} \right)^{{\frac{1}{{\sum_{{i=1}}^{{n}} w_i}}}}\]

where:

  • \(n\) is the number of indicators,

  • \(x_i\) is the value of the \(i\)-th indicator,

  • \(w_i\) is the weight assigned to the \(i\)-th indicator.

Indicators received an equal weight and domains were weighted based on BAP survey responses (see: Weighting). Aggregation was applied to aggregate indicators to domains and then to aggregate domains into a final index score. The resulting scores were then ordered and ranked such that a higher ranking indicated less behavioral health equity. Ranks were assigned for the BHEI index overall, each of the eight domains, and all 29 indicators.

Validation

Sensitivity Analysis

A sensitivity analysis was conducted to better understand and quantify the effects of the index assumptions on the BHEI rankings. Using a Monte Carlo simulation, the census tract index was calculated 1,000 times. Index assumptions were randomly varied at each iteration as follows:

  • normalization: min-max versus z-score,
  • aggregation: arithmetic mean versus geometric mean,
  • weighting: technical expert only weights, community expert only weights, or all representative weights.

The mean difference between the estimated ranks and the normative ranks for each assumption are presented in Figure 2. This demonstrates the most important source of uncertainty in the index was the normalization method, followed by aggregation. The weighting structure was effectively insignificant, having virtually no impact on the BHEI rankings. Additionally, the normalization and aggregation methods appear to have a substantial interaction effect.

Figure 2: Sensitivity analysis

Uncertainty Analysis

For each geographic unit of analysis, an uncertainty analysis was also conducted. Because the sensitivity analysis demonstrated that the weighting structure was not a significant source of uncertainty, these analyses only varied the normalization (min-max versus z-score) and aggregation methods (arithmetic versus geometric mean).

Using Monte Carlo simulations, the index was calculated 1,000 times for each geographic unit. Index assumptions for normalization and aggregation were randomly varied at each iteration. The normative rankings and the mean, 5th, and 95th percentile from the simulation ranks are shown in Figures 3-6 and Tables 11-14. In these analyses, values are reverse coded such that a higher value means conditions are more favorable for behavioral health equity.

Across all geographies, uncertainty intervals are generally lower as BHEI rankings approach minimum and maximum values, suggesting the BHEI may be most robust at the top and bottom quartiles. Additionally, uncertainty typically decreased as the geographic units increased in size, indicating a trade-off between more granular spatial resolution and more robust estimates.

Plots

SRA

Figure 3: Uncertainty analysis by SRA

Census Tract

Figure 4: Uncertainty analysis by census tract

ZCTA

Figure 5: Uncertainty analysis by ZCTA

HHSA

Figure 6: Uncertainty analysis by HHSA

Data Tables

SRA

Table 11: Uncertainty analysis by SRA

Census Tract

Table 12: Uncertainty analysis by census tract

ZCTA

Table 13: Uncertainty analysis by ZCTA

HHSA

Table 14: Uncertainty analysis by HHSA

Comparisons to HPI

The Healthy Places Index (HPI) applies a social determinants framework to rank communities in California based on social determinants of health equity (Delaney et al., 2018). The BHEI was informed by the HPI approach but focuses explicitly on predicting behavioral health equity in San Diego County. Only eight of the 29 indicators included in the BHEI are also included in the HPI and most of the assumptions around the index construction differ (e.g., weighting, aggregation, data treatment, imputation). While it is expected that the HPI and BHEI will be correlated, the assumption is that the BHEI will more accurately identify neighborhoods at risk for behavioral health inequity in San Diego County than the HPI.

It is difficult to draw comparisons between the HPI and BHEI because the BHEI uses 2020 census boundaries (n = 727) while the HPI uses 2010 boundaries (n = 628). There were only 524 census tract identifiers that appeared in both the BHEI and HPI geographic boundary files.

Analyses conducted on the subset of overlapping geographies suggest that although the BHEI and HPI were correlated (Pearson Correlation Coefficient = .90), there were important differences in rankings (Figure 7). Concordance between the HPI and BHEI was lowest in the second and third quartiles (Table 15). Nearly one in four (23%) of BHEI census tract geographies with rankings in the top quartile were not ranked in the top quartile by the HPI (Table 16). These analyses suggest that although the indices are correlated, the BHEI is identifying unique areas of need relative to the HPI.

For all analyses, to align with the BHEI, the HPI scores were reversed so that a higher score indicates less healthy conditions.

Scatterplot

Figure 7: Scatterplot of index rankings estimated from the HPI versus the BHEI

Concordance by Quartile

Table 15: Concordance by quartile for rankings estimated from the HPI versus the BHEI

Quartile HPI Total
1 2 3 4
Quartile BHEI
    1 98 (80%) 24 (20%) 0 (0%) 0 (0%) 122 (100%)
    2 29 (23%) 68 (54%) 25 (20%) 4 (3.2%) 126 (100%)
    3 2 (1.6%) 27 (22%) 80 (65%) 14 (11%) 123 (100%)
    4 0 (0%) 0 (0%) 34 (22%) 119 (78%) 153 (100%)
Total 129 (25%) 119 (23%) 139 (27%) 137 (26%) 524 (100%)

Concordance by Top Quartile

Table 16: Concordance by top quartile for rankings estimated from the HPI versus the BHEI

Top Quartile HPI Total
No Yes
Top Quartile BHEI
    No 353 (95%) 18 (4.9%) 371 (100%)
    Yes 34 (22%) 119 (78%) 153 (100%)
Total 387 (74%) 137 (26%) 524 (100%)

Results

Univariate Statistics

Univariate statistics for each of the BHEI indicators by geographic unit of analysis are presented in Tables 17-20. These statistics were calculated prior to data treatment or missing data imputation.

Census Tract

Table 17: Univariate statistics for BHEI indicators by census tract

Variable Stats / Values Freqs (% of Valid) Graph Missing
Alcohol outlets  [numeric]
Mean (sd) : 41.5 (31.6)
min ≤ med ≤ max:
0 ≤ 34.2 ≤ 100
IQR (CV) : 52.5 (0.8)
413 distinct values 19 (2.6%)
Below 200% FPL [numeric]
Mean (sd) : 24.9 (15.1)
min ≤ med ≤ max:
0 ≤ 21.7 ≤ 87.9
IQR (CV) : 20.3 (0.6)
394 distinct values 0 (0.0%)
Dental visit [numeric]
Mean (sd) : 66.5 (9.1)
min ≤ med ≤ max:
37.9 ≤ 68.3 ≤ 82.1
IQR (CV) : 12.4 (0.1)
290 distinct values 0 (0.0%)
Depression [numeric]
Mean (sd) : 17.4 (1.8)
min ≤ med ≤ max:
10.9 ≤ 17.5 ≤ 23.5
IQR (CV) : 2.3 (0.1)
89 distinct values 0 (0.0%)
Did not graduate HS [numeric]
Mean (sd) : 11.6 (10.6)
min ≤ med ≤ max:
0 ≤ 8.3 ≤ 55.8
IQR (CV) : 12.9 (0.9)
283 distinct values 0 (0.0%)
Disability [numeric]
Mean (sd) : 10 (4.2)
min ≤ med ≤ max:
0.4 ≤ 9.6 ≤ 29.6
IQR (CV) : 5.4 (0.4)
175 distinct values 0 (0.0%)
Doctor visit [numeric]
Mean (sd) : 68.8 (2.9)
min ≤ med ≤ max:
58.2 ≤ 68.7 ≤ 85.5
IQR (CV) : 3.2 (0)
136 distinct values 0 (0.0%)
ED discharge for MBD [numeric]
Mean (sd) : 4954.6 (2568.1)
min ≤ med ≤ max:
31 ≤ 4630 ≤ 19542
IQR (CV) : 3191.5 (0.5)
504 distinct values 0 (0.0%)
Income at 35 [numeric]
Mean (sd) : 44460.2 (9830.6)
min ≤ med ≤ max:
16223 ≤ 44027 ≤ 72793
IQR (CV) : 13000.5 (0.2)
626 distinct values 1 (0.1%)
IP discharge for MBD [numeric]
Mean (sd) : 3367.2 (1316.7)
min ≤ med ≤ max:
136 ≤ 3233 ≤ 10372
IQR (CV) : 1459.5 (0.4)
493 distinct values 0 (0.0%)
Limited English [numeric]
Mean (sd) : 12.7 (9.9)
min ≤ med ≤ max:
0 ≤ 9.7 ≤ 53.8
IQR (CV) : 12.8 (0.8)
286 distinct values 0 (0.0%)
Mean income [numeric]
Mean (sd) : 119195.1 (49839.7)
min ≤ med ≤ max:
34564 ≤ 111122 ≤ 380300
IQR (CV) : 55496.5 (0.4)
726 distinct values 0 (0.0%)
No automobile [numeric]
Mean (sd) : 5.4 (5.6)
min ≤ med ≤ max:
0 ≤ 3.5 ≤ 46.9
IQR (CV) : 6.1 (1.1)
175 distinct values 0 (0.0%)
No computer [numeric]
Mean (sd) : 3.6 (3.5)
min ≤ med ≤ max:
0 ≤ 2.7 ≤ 23.7
IQR (CV) : 4 (1)
124 distinct values 0 (0.0%)
No internet [numeric]
Mean (sd) : 7 (5.5)
min ≤ med ≤ max:
0 ≤ 5.8 ≤ 43.6
IQR (CV) : 6.7 (0.8)
190 distinct values 0 (0.0%)
Overcrowded [numeric]
Mean (sd) : 7.2 (7.3)
min ≤ med ≤ max:
0 ≤ 4.8 ≤ 41.4
IQR (CV) : 7.9 (1)
221 distinct values 0 (0.0%)
Owner cost burden [numeric]
Mean (sd) : 32.6 (11.6)
min ≤ med ≤ max:
0 ≤ 31.4 ≤ 100
IQR (CV) : 13.9 (0.4)
334 distinct values 4 (0.6%)
Park access [numeric]
Mean (sd) : 82.9 (26.7)
min ≤ med ≤ max:
0 ≤ 99.8 ≤ 100
IQR (CV) : 27.8 (0.3)
247 distinct values 0 (0.0%)
Poor mental health [numeric]
Mean (sd) : 14.3 (2.2)
min ≤ med ≤ max:
7.8 ≤ 14.3 ≤ 22.7
IQR (CV) : 3 (0.2)
104 distinct values 0 (0.0%)
Poor physical health [numeric]
Mean (sd) : 9.3 (2.4)
min ≤ med ≤ max:
4.3 ≤ 8.9 ≤ 16.9
IQR (CV) : 3.2 (0.3)
108 distinct values 0 (0.0%)
Renter cost burden [numeric]
Mean (sd) : 52.7 (15.7)
min ≤ med ≤ max:
0 ≤ 53.2 ≤ 100
IQR (CV) : 20.2 (0.3)
411 distinct values 1 (0.1%)
Residential diversity [numeric]
Mean (sd) : 0.5 (0.1)
min ≤ med ≤ max:
0.1 ≤ 0.5 ≤ 0.8
IQR (CV) : 0.2 (0.2)
384 distinct values 0 (0.0%)
Single-parent [numeric]
Mean (sd) : 27 (16.9)
min ≤ med ≤ max:
0 ≤ 25.5 ≤ 100
IQR (CV) : 21.8 (0.6)
405 distinct values 1 (0.1%)
Smoker [numeric]
Mean (sd) : 11.2 (3.1)
min ≤ med ≤ max:
4.8 ≤ 10.8 ≤ 22.2
IQR (CV) : 4.1 (0.3)
137 distinct values 0 (0.0%)
Unemployed [numeric]
Mean (sd) : 6.4 (3.9)
min ≤ med ≤ max:
0 ≤ 5.6 ≤ 32.5
IQR (CV) : 4.8 (0.6)
151 distinct values 0 (0.0%)
Uninsured [numeric]
Mean (sd) : 7.4 (5.5)
min ≤ med ≤ max:
0 ≤ 6.3 ≤ 33.8
IQR (CV) : 7 (0.7)
198 distinct values 0 (0.0%)
Voting [numeric]
Mean (sd) : 81.3 (8)
min ≤ med ≤ max:
55.1 ≤ 83.4 ≤ 93.9
IQR (CV) : 10.8 (0.1)
262 distinct values 0 (0.0%)

Generated by summarytools 1.0.1 (R version 4.1.2)
2023-06-30

ZCTA

Table 18: Univariate statistics for BHEI indicators by ZCTA

Variable Stats / Values Freqs (% of Valid) Graph Missing
Alcohol outlets  [numeric]
Mean (sd) : 30.1 (23.8)
min ≤ med ≤ max:
0 ≤ 26 ≤ 98.1
IQR (CV) : 29.9 (0.8)
84 distinct values 0 (0.0%)
Below 200% FPL [numeric]
Mean (sd) : 24.5 (11)
min ≤ med ≤ max:
6 ≤ 21.3 ≤ 60.2
IQR (CV) : 12.2 (0.4)
84 distinct values 0 (0.0%)
Dental visit [numeric]
Mean (sd) : 67.2 (6.8)
min ≤ med ≤ max:
47.2 ≤ 68.7 ≤ 80.2
IQR (CV) : 9.7 (0.1)
79 distinct values 0 (0.0%)
Depression [numeric]
Mean (sd) : 17.6 (1.4)
min ≤ med ≤ max:
14 ≤ 17.8 ≤ 20.1
IQR (CV) : 1.7 (0.1)
48 distinct values 0 (0.0%)
Did not graduate HS [numeric]
Mean (sd) : 10.2 (7.8)
min ≤ med ≤ max:
0 ≤ 8.1 ≤ 38.5
IQR (CV) : 9.8 (0.8)
75 distinct values 0 (0.0%)
Disability [numeric]
Mean (sd) : 10.1 (4.6)
min ≤ med ≤ max:
2.9 ≤ 10.1 ≤ 44.8
IQR (CV) : 3.8 (0.5)
59 distinct values 0 (0.0%)
Doctor visit [numeric]
Mean (sd) : 69.2 (2.5)
min ≤ med ≤ max:
58.2 ≤ 69.2 ≤ 75.9
IQR (CV) : 2.8 (0)
51 distinct values 0 (0.0%)
ED discharge for MBD [numeric]
Mean (sd) : 4647.2 (2122.5)
min ≤ med ≤ max:
31 ≤ 4737 ≤ 14975
IQR (CV) : 2568 (0.5)
94 distinct values 0 (0.0%)
Income at 35 [numeric]
Mean (sd) : 45244.9 (7609.8)
min ≤ med ≤ max:
28672 ≤ 44529 ≤ 62533
IQR (CV) : 10498 (0.2)
95 distinct values 0 (0.0%)
IP discharge for MBD [numeric]
Mean (sd) : 3325 (1191.3)
min ≤ med ≤ max:
136 ≤ 3217 ≤ 7805
IQR (CV) : 1402 (0.4)
94 distinct values 0 (0.0%)
Limited English [numeric]
Mean (sd) : 10.3 (7.3)
min ≤ med ≤ max:
0 ≤ 8.8 ≤ 40.7
IQR (CV) : 7.8 (0.7)
76 distinct values 0 (0.0%)
Mean income [numeric]
Mean (sd) : 120779.5 (43564.6)
min ≤ med ≤ max:
41140 ≤ 112388 ≤ 316564
IQR (CV) : 44517 (0.4)
97 distinct values 0 (0.0%)
No automobile [numeric]
Mean (sd) : 4.5 (3.3)
min ≤ med ≤ max:
0 ≤ 3.9 ≤ 20.3
IQR (CV) : 3.3 (0.7)
56 distinct values 0 (0.0%)
No computer [numeric]
Mean (sd) : 3.7 (3)
min ≤ med ≤ max:
0 ≤ 3.2 ≤ 19.8
IQR (CV) : 2.8 (0.8)
54 distinct values 0 (0.0%)
No internet [numeric]
Mean (sd) : 7.9 (5.5)
min ≤ med ≤ max:
0.9 ≤ 6.6 ≤ 36.5
IQR (CV) : 5.5 (0.7)
71 distinct values 0 (0.0%)
Overcrowded [numeric]
Mean (sd) : 6.4 (5.5)
min ≤ med ≤ max:
0 ≤ 4.8 ≤ 34.3
IQR (CV) : 5.1 (0.9)
65 distinct values 0 (0.0%)
Owner cost burden [numeric]
Mean (sd) : 31.8 (5.8)
min ≤ med ≤ max:
6 ≤ 32.3 ≤ 48.3
IQR (CV) : 7 (0.2)
80 distinct values 0 (0.0%)
Park access [numeric]
Mean (sd) : 77.8 (22.8)
min ≤ med ≤ max:
3.6 ≤ 83.5 ≤ 100
IQR (CV) : 28.7 (0.3)
79 distinct values 0 (0.0%)
Poor mental health [numeric]
Mean (sd) : 14.3 (1.7)
min ≤ med ≤ max:
10.6 ≤ 14.4 ≤ 18.6
IQR (CV) : 2.2 (0.1)
55 distinct values 0 (0.0%)
Poor physical health [numeric]
Mean (sd) : 9.5 (2.2)
min ≤ med ≤ max:
5.2 ≤ 9.4 ≤ 14.5
IQR (CV) : 2.7 (0.2)
53 distinct values 0 (0.0%)
Renter cost burden [numeric]
Mean (sd) : 52.8 (13.1)
min ≤ med ≤ max:
0 ≤ 54.2 ≤ 91.9
IQR (CV) : 12.1 (0.2)
85 distinct values 0 (0.0%)
Residential diversity [numeric]
Mean (sd) : 0.6 (0.1)
min ≤ med ≤ max:
0 ≤ 0.6 ≤ 0.7
IQR (CV) : 0.1 (0.2)
83 distinct values 0 (0.0%)
Single-parent [numeric]
Mean (sd) : 27.1 (11.5)
min ≤ med ≤ max:
0 ≤ 25.7 ≤ 62.1
IQR (CV) : 12.6 (0.4)
87 distinct values 0 (0.0%)
Smoker [numeric]
Mean (sd) : 11.3 (2.6)
min ≤ med ≤ max:
6.3 ≤ 11.3 ≤ 17.3
IQR (CV) : 3.7 (0.2)
62 distinct values 0 (0.0%)
Unemployed [numeric]
Mean (sd) : 6.3 (3.2)
min ≤ med ≤ max:
0 ≤ 6.1 ≤ 27.3
IQR (CV) : 2.2 (0.5)
54 distinct values 0 (0.0%)
Uninsured [numeric]
Mean (sd) : 7.3 (4.5)
min ≤ med ≤ max:
0.8 ≤ 6.6 ≤ 27.5
IQR (CV) : 5.2 (0.6)
70 distinct values 0 (0.0%)
Voting [numeric]
Mean (sd) : 82.6 (6.4)
min ≤ med ≤ max:
61.2 ≤ 84.2 ≤ 91.3
IQR (CV) : 7.3 (0.1)
78 distinct values 0 (0.0%)

Generated by summarytools 1.0.1 (R version 4.1.2)
2023-06-30

SRA

Table 19: Univariate statistics for BHEI indicators by SRA

Variable Stats / Values Freqs (% of Valid) Graph Missing
Alcohol outlets  [numeric]
Mean (sd) : 30.2 (21.7)
min ≤ med ≤ max:
0 ≤ 26.5 ≤ 85.3
IQR (CV) : 21.8 (0.7)
40 distinct values 0 (0.0%)
Below 200% FPL [numeric]
Mean (sd) : 25.2 (9.9)
min ≤ med ≤ max:
11.6 ≤ 24.2 ≤ 50.6
IQR (CV) : 12.5 (0.4)
38 distinct values 0 (0.0%)
Dental visit [numeric]
Mean (sd) : 66.4 (6.1)
min ≤ med ≤ max:
54.4 ≤ 67 ≤ 75.9
IQR (CV) : 7.2 (0.1)
38 distinct values 0 (0.0%)
Depression [numeric]
Mean (sd) : 17.7 (1.3)
min ≤ med ≤ max:
14.5 ≤ 17.9 ≤ 20
IQR (CV) : 1.7 (0.1)
27 distinct values 0 (0.0%)
Did not graduate HS [numeric]
Mean (sd) : 10.3 (6.7)
min ≤ med ≤ max:
0 ≤ 8.9 ≤ 26.6
IQR (CV) : 8.6 (0.7)
37 distinct values 0 (0.0%)
Disability [numeric]
Mean (sd) : 10.2 (3)
min ≤ med ≤ max:
4.7 ≤ 10.1 ≤ 20.8
IQR (CV) : 3 (0.3)
31 distinct values 0 (0.0%)
Doctor visit [numeric]
Mean (sd) : 68.9 (3)
min ≤ med ≤ max:
58.2 ≤ 68.9 ≤ 75.9
IQR (CV) : 2.3 (0)
31 distinct values 0 (0.0%)
ED discharge for MBD [numeric]
Mean (sd) : 4732.4 (1980.7)
min ≤ med ≤ max:
31 ≤ 4843 ≤ 8955
IQR (CV) : 3448 (0.4)
41 distinct values 0 (0.0%)
Income at 35 [numeric]
Mean (sd) : 44785.9 (6620.3)
min ≤ med ≤ max:
32383 ≤ 45634 ≤ 57602
IQR (CV) : 9575 (0.1)
41 distinct values 0 (0.0%)
IP discharge for MBD [numeric]
Mean (sd) : 3379 (1201.3)
min ≤ med ≤ max:
136 ≤ 3247 ≤ 6135
IQR (CV) : 1637 (0.4)
41 distinct values 0 (0.0%)
Limited English [numeric]
Mean (sd) : 10.4 (6.7)
min ≤ med ≤ max:
2 ≤ 9.7 ≤ 28
IQR (CV) : 8.4 (0.6)
36 distinct values 0 (0.0%)
Mean income [numeric]
Mean (sd) : 115196.2 (31953.7)
min ≤ med ≤ max:
65045 ≤ 109046 ≤ 196259
IQR (CV) : 41176 (0.3)
41 distinct values 0 (0.0%)
No automobile [numeric]
Mean (sd) : 4.4 (2.8)
min ≤ med ≤ max:
0 ≤ 3.8 ≤ 12.3
IQR (CV) : 3.1 (0.6)
34 distinct values 0 (0.0%)
No computer [numeric]
Mean (sd) : 3.6 (2.1)
min ≤ med ≤ max:
0 ≤ 3.4 ≤ 10.7
IQR (CV) : 2 (0.6)
31 distinct values 0 (0.0%)
No internet [numeric]
Mean (sd) : 7.8 (4.4)
min ≤ med ≤ max:
0.7 ≤ 7.2 ≤ 20.1
IQR (CV) : 4.6 (0.6)
34 distinct values 0 (0.0%)
Overcrowded [numeric]
Mean (sd) : 6.3 (3.9)
min ≤ med ≤ max:
1.2 ≤ 5.4 ≤ 17.3
IQR (CV) : 4.3 (0.6)
32 distinct values 0 (0.0%)
Owner cost burden [numeric]
Mean (sd) : 33.9 (11.3)
min ≤ med ≤ max:
22.2 ≤ 33 ≤ 100
IQR (CV) : 4.4 (0.3)
32 distinct values 1 (2.4%)
Park access [numeric]
Mean (sd) : 71.5 (27.4)
min ≤ med ≤ max:
0 ≤ 79.9 ≤ 100
IQR (CV) : 32.8 (0.4)
38 distinct values 0 (0.0%)
Poor mental health [numeric]
Mean (sd) : 14.6 (1.7)
min ≤ med ≤ max:
11.7 ≤ 14.6 ≤ 19.2
IQR (CV) : 2 (0.1)
30 distinct values 0 (0.0%)
Poor physical health [numeric]
Mean (sd) : 9.5 (2.1)
min ≤ med ≤ max:
5.2 ≤ 9.7 ≤ 14.5
IQR (CV) : 3.3 (0.2)
33 distinct values 0 (0.0%)
Renter cost burden [numeric]
Mean (sd) : 52.7 (10.9)
min ≤ med ≤ max:
28 ≤ 52.7 ≤ 83.5
IQR (CV) : 11.9 (0.2)
37 distinct values 0 (0.0%)
Residential diversity [numeric]
Mean (sd) : 0.6 (0.1)
min ≤ med ≤ max:
0.4 ≤ 0.6 ≤ 0.7
IQR (CV) : 0.1 (0.2)
39 distinct values 0 (0.0%)
Single-parent [numeric]
Mean (sd) : 25.5 (8.6)
min ≤ med ≤ max:
8.5 ≤ 23 ≤ 48.1
IQR (CV) : 11.4 (0.3)
36 distinct values 0 (0.0%)
Smoker [numeric]
Mean (sd) : 11.5 (2.3)
min ≤ med ≤ max:
7.7 ≤ 11.9 ≤ 17.3
IQR (CV) : 3.1 (0.2)
32 distinct values 0 (0.0%)
Unemployed [numeric]
Mean (sd) : 6.2 (2.5)
min ≤ med ≤ max:
1.2 ≤ 5.8 ≤ 14.6
IQR (CV) : 2.8 (0.4)
32 distinct values 0 (0.0%)
Uninsured [numeric]
Mean (sd) : 7 (3.6)
min ≤ med ≤ max:
1.1 ≤ 6.1 ≤ 19.4
IQR (CV) : 4.9 (0.5)
36 distinct values 0 (0.0%)
Voting [numeric]
Mean (sd) : 81.5 (7)
min ≤ med ≤ max:
61.3 ≤ 84.1 ≤ 90.3
IQR (CV) : 6.2 (0.1)
38 distinct values 0 (0.0%)

Generated by summarytools 1.0.1 (R version 4.1.2)
2023-06-30

HHSA

Table 20: Univariate statistics for BHEI indicators by HHSA

Variable Stats / Values Freqs (% of Valid) Missing
Geography [character]
1. CENTRAL
2. EAST
3. NORTH CENTRAL
4. NORTH COASTAL
5. NORTH INLAND
6. SOUTH
1(16.7%)
1(16.7%)
1(16.7%)
1(16.7%)
1(16.7%)
1(16.7%)
0 (0.0%)
Alcohol outlets  [numeric]
Mean (sd) : 39.2 (15.8)
min ≤ med ≤ max:
24 ≤ 37.5 ≤ 67.1
IQR (CV) : 15.5 (0.4)
24.00:1(16.7%)
25.10:1(16.7%)
34.60:1(16.7%)
40.40:1(16.7%)
43.90:1(16.7%)
67.10:1(16.7%)
0 (0.0%)
Below 200% FPL [numeric]
Mean (sd) : 25.6 (5.7)
min ≤ med ≤ max:
19 ≤ 24.9 ≤ 35
IQR (CV) : 5.2 (0.2)
19.00:1(16.7%)
22.30:1(16.7%)
22.40:1(16.7%)
27.40:1(16.7%)
27.60:1(16.7%)
35.00:1(16.7%)
0 (0.0%)
Dental visit [numeric]
Mean (sd) : 65.8 (4.6)
min ≤ med ≤ max:
59.6 ≤ 66.8 ≤ 71.5
IQR (CV) : 6.1 (0.1)
59.60:1(16.7%)
61.60:1(16.7%)
65.20:1(16.7%)
68.30:1(16.7%)
68.70:1(16.7%)
71.50:1(16.7%)
0 (0.0%)
Depression [numeric]
Mean (sd) : 17.4 (1)
min ≤ med ≤ max:
16.2 ≤ 17.4 ≤ 19
IQR (CV) : 0.9 (0.1)
16.20:1(16.7%)
16.60:1(16.7%)
17.30:1(16.7%)
17.50:1(16.7%)
17.70:1(16.7%)
19.00:1(16.7%)
0 (0.0%)
Did not graduate HS [numeric]
Mean (sd) : 12.1 (4.9)
min ≤ med ≤ max:
5.3 ≤ 10.8 ≤ 18.9
IQR (CV) : 5 (0.4)
5.30:1(16.7%)
10.20:2(33.3%)
11.40:1(16.7%)
16.50:1(16.7%)
18.90:1(16.7%)
0 (0.0%)
Disability [numeric]
Mean (sd) : 10 (1.4)
min ≤ med ≤ max:
8.4 ≤ 9.8 ≤ 12.6
IQR (CV) : 1 (0.1)
8.40:1(16.7%)
9.20:1(16.7%)
9.50:1(16.7%)
10.20:1(16.7%)
10.30:1(16.7%)
12.60:1(16.7%)
0 (0.0%)
Doctor visit [numeric]
Mean (sd) : 68.6 (0.7)
min ≤ med ≤ max:
67.7 ≤ 68.4 ≤ 69.4
IQR (CV) : 1.1 (0)
67.70:1(16.7%)
68.00:1(16.7%)
68.30:1(16.7%)
68.50:1(16.7%)
69.40:2(33.3%)
0 (0.0%)
ED discharge for MBD [numeric]
Mean (sd) : 4942.7 (1815.6)
min ≤ med ≤ max:
3236 ≤ 4340.5 ≤ 7611
IQR (CV) : 2619.2 (0.4)
3236:1(16.7%)
3391:1(16.7%)
4121:1(16.7%)
4560:1(16.7%)
6737:1(16.7%)
7611:1(16.7%)
0 (0.0%)
Income at 35 [numeric]
Mean (sd) : 44186.7 (5187.3)
min ≤ med ≤ max:
36363 ≤ 45515 ≤ 49912
IQR (CV) : 6597.2 (0.1)
36363:1(16.7%)
39629:1(16.7%)
45156:1(16.7%)
45874:1(16.7%)
48186:1(16.7%)
49912:1(16.7%)
0 (0.0%)
IP discharge for MBD [numeric]
Mean (sd) : 3358.7 (935.9)
min ≤ med ≤ max:
2409 ≤ 3014 ≤ 4720
IQR (CV) : 1296.8 (0.3)
2409:1(16.7%)
2708:1(16.7%)
2789:1(16.7%)
3239:1(16.7%)
4287:1(16.7%)
4720:1(16.7%)
0 (0.0%)
Limited English [numeric]
Mean (sd) : 13.3 (4.8)
min ≤ med ≤ max:
9.5 ≤ 11.2 ≤ 21.1
IQR (CV) : 6.1 (0.4)
9.50:1(16.7%)
9.60:1(16.7%)
10.50:1(16.7%)
11.80:1(16.7%)
17.30:1(16.7%)
21.10:1(16.7%)
0 (0.0%)
Mean income [numeric]
Mean (sd) : 116453.2 (19129.3)
min ≤ med ≤ max:
94638 ≤ 114838.5 ≤ 139693
IQR (CV) : 29634.5 (0.2)
94638:1(16.7%)
101858:1(16.7%)
102062:1(16.7%)
127615:1(16.7%)
132853:1(16.7%)
139693:1(16.7%)
0 (0.0%)
No automobile [numeric]
Mean (sd) : 5.5 (2.2)
min ≤ med ≤ max:
3.7 ≤ 4.9 ≤ 9.6
IQR (CV) : 1.5 (0.4)
3.70:1(16.7%)
4.20:1(16.7%)
4.30:1(16.7%)
5.60:1(16.7%)
5.70:1(16.7%)
9.60:1(16.7%)
0 (0.0%)
No computer [numeric]
Mean (sd) : 3.7 (1.1)
min ≤ med ≤ max:
2.3 ≤ 3.7 ≤ 5
IQR (CV) : 1.5 (0.3)
2.30:1(16.7%)
2.90:1(16.7%)
3.10:1(16.7%)
4.20:1(16.7%)
4.50:1(16.7%)
5.00:1(16.7%)
0 (0.0%)
No internet [numeric]
Mean (sd) : 7.1 (2)
min ≤ med ≤ max:
4.3 ≤ 7.1 ≤ 9.4
IQR (CV) : 3 (0.3)
4.30:1(16.7%)
5.70:1(16.7%)
6.20:1(16.7%)
7.90:1(16.7%)
9.20:1(16.7%)
9.40:1(16.7%)
0 (0.0%)
Overcrowded [numeric]
Mean (sd) : 7.3 (2.7)
min ≤ med ≤ max:
4.2 ≤ 6.4 ≤ 11.7
IQR (CV) : 2.9 (0.4)
4.20:1(16.7%)
5.70:1(16.7%)
5.80:1(16.7%)
7.00:1(16.7%)
9.20:1(16.7%)
11.70:1(16.7%)
0 (0.0%)
Owner cost burden [numeric]
Mean (sd) : 32.1 (2)
min ≤ med ≤ max:
28.4 ≤ 32.5 ≤ 34.2
IQR (CV) : 0.8 (0.1)
28.40:1(16.7%)
32.00:1(16.7%)
32.30:1(16.7%)
32.60:1(16.7%)
33.00:1(16.7%)
34.20:1(16.7%)
0 (0.0%)
Park access [numeric]
Mean (sd) : 81.3 (12.3)
min ≤ med ≤ max:
65.5 ≤ 82.4 ≤ 93.3
IQR (CV) : 19.7 (0.2)
65.50:1(16.7%)
71.80:1(16.7%)
74.20:1(16.7%)
90.70:1(16.7%)
92.60:1(16.7%)
93.30:1(16.7%)
0 (0.0%)
Poor mental health [numeric]
Mean (sd) : 14.5 (0.9)
min ≤ med ≤ max:
13.2 ≤ 14.4 ≤ 15.7
IQR (CV) : 1.2 (0.1)
13.20:1(16.7%)
13.90:1(16.7%)
14.30:1(16.7%)
14.50:1(16.7%)
15.40:1(16.7%)
15.70:1(16.7%)
0 (0.0%)
Poor physical health [numeric]
Mean (sd) : 9.4 (1.2)
min ≤ med ≤ max:
7.4 ≤ 9.7 ≤ 10.5
IQR (CV) : 1.2 (0.1)
7.40:1(16.7%)
8.80:1(16.7%)
9.30:1(16.7%)
10.10:1(16.7%)
10.20:1(16.7%)
10.50:1(16.7%)
0 (0.0%)
Renter cost burden [numeric]
Mean (sd) : 55 (3.2)
min ≤ med ≤ max:
50 ≤ 55.3 ≤ 58.2
IQR (CV) : 4.1 (0.1)
50.00:1(16.7%)
53.40:1(16.7%)
53.90:1(16.7%)
56.80:1(16.7%)
57.90:1(16.7%)
58.20:1(16.7%)
0 (0.0%)
Residential diversity [numeric]
Mean (sd) : 0.6 (0)
min ≤ med ≤ max:
0.6 ≤ 0.6 ≤ 0.7
IQR (CV) : 0 (0.1)
0.58  :1(16.7%)
0.60 !:1(16.7%)
0.61 !:1(16.7%)
0.63 !:1(16.7%)
0.64 !:1(16.7%)
0.71  :1(16.7%)
! rounded
0 (0.0%)
Single-parent [numeric]
Mean (sd) : 27.1 (5.6)
min ≤ med ≤ max:
21 ≤ 26.5 ≤ 36.6
IQR (CV) : 5.4 (0.2)
21.00:1(16.7%)
23.00:1(16.7%)
24.70:1(16.7%)
28.40:1(16.7%)
29.00:1(16.7%)
36.60:1(16.7%)
0 (0.0%)
Smoker [numeric]
Mean (sd) : 11.4 (1.5)
min ≤ med ≤ max:
9.1 ≤ 11.2 ≤ 13.2
IQR (CV) : 1.9 (0.1)
9.10:1(16.7%)
10.60:1(16.7%)
10.90:1(16.7%)
11.40:1(16.7%)
13.00:1(16.7%)
13.20:1(16.7%)
0 (0.0%)
Unemployed [numeric]
Mean (sd) : 6.4 (1.4)
min ≤ med ≤ max:
5 ≤ 6.2 ≤ 8.5
IQR (CV) : 1.9 (0.2)
5.00:1(16.7%)
5.20:1(16.7%)
5.40:1(16.7%)
6.90:1(16.7%)
7.20:1(16.7%)
8.50:1(16.7%)
0 (0.0%)
Uninsured [numeric]
Mean (sd) : 7.6 (1.9)
min ≤ med ≤ max:
5.1 ≤ 7.5 ≤ 10.3
IQR (CV) : 2.3 (0.2)
5.10:1(16.7%)
6.10:1(16.7%)
7.30:1(16.7%)
7.70:1(16.7%)
9.00:1(16.7%)
10.30:1(16.7%)
0 (0.0%)
Voting [numeric]
Mean (sd) : 81.7 (4.9)
min ≤ med ≤ max:
75.4 ≤ 83.4 ≤ 86
IQR (CV) : 8.2 (0.1)
75.40:1(16.7%)
76.20:1(16.7%)
81.50:1(16.7%)
85.40:1(16.7%)
85.80:1(16.7%)
86.00:1(16.7%)
0 (0.0%)

Generated by summarytools 1.0.1 (R version 4.1.2)
2023-06-30

BHEI Structure

The final BHEI structure is presented in Figure 8. The domain and indicator boxes are sized to reflect the effective weights for the index. Becker (2022) defines effective weights as the: “final weights of each indicator in the index, as a result of the indicator weights, the parent aggregate weights, and the structure of the index.”

Figure 8: BHEI structure and effective weights

Correlations

The interpretation of correlation structures in indices is debated. Some experts maintain that an index’s elements should be positively correlated since they are meant to measure the same general concept and are more likely to transfer information to the index when the assumption of positive correlations holds (Linden et al., 2021). Other researchers advocate for the inclusion of negatively or non-significantly correlated indicators since these will provide unique information instead of “double counting” correlated indicators (Nardo et al, 2008).

Given that there is not clear consensus about whether indicators should be positively or negatively correlated, Linden et al (2021) recommend a “balanced approach” whereby fewer indicators should be aggregated if they are negatively or non-significantly correlated, while more indicators can be aggregated if correlations are positive.

Pearson Correlation Coefficients between and across different levels of the index are presented in Figures 9-13. In these tables, correlations that did not achieve statistical significance at the p<.05 level appear as gray. Indicators included in Table 9 were rescaled prior to this analysis.

The components of the index were generally positively and significantly correlated, which suggests the BHEI can effectively aggregate the relatively large number of indicators (i.e., 29). The most obvious exception is “Self-Reported Unmet Need for Mental Health Treatment” which had weak or negative correlations with some of the composite parts of the index. The development team felt this indicator contributed important and unique information to the BHEI and so the indicator was retained.

Domains to Index

Figure 9: Correlation structure between the BHEI index and domains

Indicators to Index

Figure 10: Correlation structure between the BHEI index and indicators

Domains to Domains

Figure 11: Correlation structure between the BHEI domains

Domains to Indicators

Figure 12: Correlation structure between the BHEI domains and indicators

Indicators to Indicators

Figure 13: Correlation structure between the BHEI indicators

BHEI Ranks

Tables 21-24 show the final BHEI rankings for the overall index, the eight BHEI domains, and the 29 indicators by each geographic unit of analysis. Neighborhoods with higher BHEI scores are estimated to have more behavioral health inequity (i.e., to have less access to the resources, opportunities, and conditions that promote behavioral health than neighborhoods with lower BHEI scores). Indicators sourced from CHIS Neighborhood Edition have been suppressed to comply with data use agreements.

Census Tract

Table 21: BHEI census tract ranks by index, domain, and indicator

ZCTA

Table 22: BHEI ZCTA ranks by index, domain, and indicator

SRA

Table 23: BHEI SRA ranks by index, domain, and indicator

HHSA

Table 24: BHEI HHSA ranks by index, domain, and indicator

BHEI Maps

Figures 14-17 map the BHEI rankings by geographic unit of analysis. Darker colors indicate higher ranks (i.e., areas that are less likely to have access to the conditions and opportunities that promote behavioral health equity).

Census Tract

Figure 14: Map of BHEI ranks by census tract

ZCTA

Figure 15: Map of BHEI ranks by ZCTA

SRA

Figure 16: Map of BHEI ranks by SRA

HHSA

Figure 17: Map of BHEI ranks by HHSA

Total Population

The total estimated population by BHEI quartile for the census tract, ZCTA, and SRA indices are presented in Table 25. Quartile data are not presented for HHSAs since there are only six regions. Higher quartiles identify areas that are less likely to have access to the conditions and opportunities that promote behavioral health equity. Populations do not sum to the same total across geographic units because some census tract and ZCTA geographies were excluded and some ZCTAs cross county boundaries.

Table 25: Estimated population size by BHEI quartile and geographic unit

Sociodemographic Profiles

To assess the face validity of the BHEI, sociodemographic profiles were compared for geographies with BHEI rankings in the top and bottom quartiles (Figures 18-20 and Tables 26-28). As hypothesized, populations known to experience behavioral health inequities were more highly concentrated in regions with BHEI rankings in the top quartile. This relationship was observed across census tract, ZCTA, and SRA geographies. These results further confirm the validity of the BHEI and its ability to identify neighborhoods or regions at increased risk for behavioral health inequities. Outcomes with an asterisk are included as indicators in the BHEI and cannot independently assess validity.

Plots

Census Tract

Figure 18: Sociodemographic profiles by top and bottom BHEI census tract quartile

ZCTA

Figure 19: Sociodemographic profiles by top and bottom BHEI ZCTA quartile

SRA

Figure 20: Sociodemographic profiles by top and bottom BHEI SRA quartile

Data Tables

In Tables 26-28 the Relative Risk (RR) is calculated as the ratio between the percentage in Quartile Four and Quartile One. A RR greater than one indicates the percentage is greater among the Quartile Four population than the Quartile One population.

Census Tract

Table 26: Sociodemographic profiles by top and bottom BHEI census tract quartile

ZCTA

Table 27: Sociodemographic profiles by top and bottom BHEI ZCTA quartile

SRA

Table 28: Sociodemographic profiles by top and bottom BHEI SRA quartile

Discussion

Interpreting the BHEI

Neighborhoods with higher BHEI scores are relatively less likely to have access to the resources, opportunities, and conditions that promote behavioral health than neighborhoods with lower BHEI scores. In this way, a higher score may be an indicator of systemic inequities in policies, laws, and service provision. Areas with higher BHEI scores may benefit from behavioral health service enhancements or quality improvement efforts.

Using the BHEI

The BHEI has been programmed into the Community Experience Partnership: Service Planning Tool. The application generates interactive maps which allow users to explore BHEI rankings across different geographies, weight the index by selected target populations, and generate parameterized summary reports to gain a better understanding of sociodemographic conditions in selected areas. Once target neighborhoods are identified, next steps would likely include engaging community representatives from these areas to gain a better understanding of the local strengths, needs, and resources.

Strengths and Limitations

The BHEI has numerous strengths. The indicators, domains, and weights were developed collaboratively with community partners and SMEs to reflect the most salient predictors of behavioral health equity in San Diego County. The index includes numerous indicators not available in other indices, leverages state of the art packages and advanced statistical methods for construction and validation (v1.1.4; Becker; 2022), and utilizes new 2020 U.S. Census geographic boundaries for increased precision. Sensitivity and uncertainty analyses confirm the robustness of the rankings, especially at the top and bottom quartiles. Additionally, the index has demonstrated excellent face validity through post-hoc concordance and subgroup analyses. Findings from this report suggest the BHEI may be a valuable tool to inform culturally responsive, data-informed outreach efforts to previously underserved areas. The BHEI may be further improved through community review and feedback.

Despite these strengths, there are important limitations to consider when using the BHEI. Most importantly, the BHEI is not intended to be applied or interpreted without context. The ranks do not reflect the strengths, values, or priorities of neighborhoods or regions and the individuals who live there. While the BHEI can help users identify neighborhoods that may benefit from service enhancements and quality improvement efforts, final decisions about needs, policy, and resourcing would require additional analyses and community outreach.

Additional limitations are listed below:

  1. BHEI ranks are relative; they can show that one area is more or less equitable than another area but not by how much.
  2. Geographic aggregation may conceal areas of need. For instance, a SRA may have a relatively low BHEI ranking despite including some neighborhoods at very high risk for behavioral health inequity. This highlights the importance of engaging local stakeholders and community experts when interpreting the BHEI.
  3. Data sources are not available for all indicators considered important causes of behavioral health equity in San Diego County (e.g., indicators of systemic racism, crime, etc.).
  4. Although the most recently available data were always used, some sources have not been updated for a number of years and may not reflect current conditions.
  5. The geographic reallocation methods assume outcomes are equally distributed across geographic areas, which may introduce inaccuracies.
  6. The BHEI indicators are mainly composed of social determinants of behavioral health. Most of these “upriver” correlates of behavioral health (e.g., poverty and education) will not be impacted by enhancements in behavioral health service provision or outreach. As such, the BHEI cannot be used to evaluate the efficacy of targeted behavioral health interventions or policy changes.

References

Becker, J., Starnberger, M., Schneller, O., & Lederer, M. (2022). COINr: An R package for developing composite indicators. Journal of Open Source Software, 7(78), 4567. doi: 10.21105/joss.04567

Benjamin, D. J., Heffetz, O., Kimball, M. S., & Szembrot, N. (2014). Beyond happiness and satisfaction: Toward well-being indices based on stated preference. American Economic Review, 104(9), 2698-2735.

California Department of Public Health, California Department of Health Care Access and Information (HCAI), Emergency Department Data, Patient Discharge Data: Version 4/2023 [Data set]. Prepared by County of San Diego, Health and Human Services Agency, Behavioral Health Services, Population Health Unit. Versions 4/12/2023.

California Healthy Places Index (HPI 3.0). Public Health Alliance of Southern California. Retrieved [May 2023], from https://www.healthyplacesindex.org

Decancq, K., & Lugo, M. A. (2013). Weights in multidimensional indices of wellbeing: An overview. Econometric Reviews, 32(1), 7-34.

Delaney, T., Dominie, W., Dowling, H., & Maizlish, N. (2018). Healthy Places Index: Technical Documentation. Los Angeles, CA: Public Health Alliance of Southern California. Retrieved [May 2023], from https://www.healthyplacesindex.org

Esri. (2022). An ESRI Technical Paper: Methodology statement 2016-2020 American Community Survey. Retrieved [May 2023] from https://downloads.esri.com/Support/downloads/other_/2022/2022_USA_ESRI_American_Community_Survey.pdf

Greco, S., Ishizaka, A., Tasiou, M., et al. (2019). On the Methodological Framework of Composite Indices: A Review of the Issues of Weighting, Aggregation, and Robustness. Social Indicators Research, 141(1), 61-94. doi: 10.1007/s11205-017-1832-9

Healthy People 2030, U.S. Department of Health and Human Services, Office of Disease Prevention and Health Promotion. Retrieved [May 20, 2022], from https://health.gov/healthypeople/objectives-and-data/social-determinants-health

Kim, H. Y. (2013). Statistical notes for clinical researchers: Assessing normal distribution (2) using skewness and kurtosis. Restorative Dentistry & Endodontics, 38(1), 52-54. doi: 10.5395/rde.2013.38.1.52

Ma, M. San Diego Association of Governments Population and Housing Estimates: Versions 2021. [dataset]. SANDAG.

Matthews, SA., Stoltz, DS., Perkins, TA (2020). segregation: Entropy-based Segregation Measures. R package version 1.4.1. https://CRAN.R-project.org/package=segregation

Manson, S., Schroeder, J., Van Riper, D., Kugler, T, Ruggles, S. IPUMS National Historical Geographic Information System: Version 16.0 [dataset]. Minneapolis, MN: IPUMS. 2021. http://doi.org/10.18128/D050.V16.0

UC San Diego’s Homelessness Hub: RTFH - Point-in-Time Counts, 2018: Version 03/2022. [dataset]. (https://UC San Diegoonline.maps.arcgis.com/home/item.html?id=fcea861c6a3b40fcadd67e8fb3c00c

Appendix A

Information about the indicators that make up the BHEI, including data sources, most recent year(s) the data were produced, and technical definitions, are included in Table 29.

Table 29: BHEI Indicator Codebook